C-EB/Fine-Tuning-mBART-for-Darija-to-English-Translation

Domaine:

natural language processing

Type de record:

model

Créateur:

C-EB

Hôte:

Fine-tuning mBART for Moroccan Darija–English translation using a custom parallel dataset, with a deployed Streamlit demo and publicly accessible model on Hugging Face Hub. # Fine-Tuning mBART for Moroccan Darija ↔ English Translation ## Project Overview This project focuses on fine-tuning the multilingual mBART model (`facebook/mbart-large-50-many-to-many-mmt`) for translation between Moroccan Darija and English. Moroccan Darija, a spoken dialect of Arabic, is underrepresented in machine translation systems. This work aims to address that by adapting a large multilingual transformer model using a custom parallel corpus. Alongside the training pipeline, a simple web-based translation interface was built using Streamlit and deployed to **Hugging Face Spaces**. The trained model was also uploaded to the **Hugging Face Hub** to make it publicly accessible and easily integrated into other applications. ## Goals * Fine-tune mBART on a custom Moroccan Darija–English parallel dataset. * Improve translation quality for Darija, a low-resource language. * Provide an easy-to-use web app for real-time translation. * Host the model and demo openly for the community. ## Tools & Libraries Used * **Transformers (Hugging Face)** – Tokenizer, model, training loop * **Datasets (Hugging Face)** – Dataset handling and preprocessing * **PyTorch** – Training backend * **scikit-learn** – Dataset splitting * **pandas / numpy** – Data manipulation * **sacrebleu** – Evaluation metric * **Streamlit** – Interactive web interface * **Hugging Face Hub / Spaces** – Model and app deployment ## Workflow ### 1. Environment Setup Install necessary packages: ```bash !pip install transformers datasets sacrebleu streamlit ``` ### 2. Data Preparation * Load CSV dataset with parallel text (`darija`, `english`). * Split into training, validation, and test sets. * Convert to `DatasetDict` format for Hugging Face training. ### 3. Tokenization * Use mBART tokenizer with language codes: * `ar_AR` for Darija * `en_XX` for English * Apply padding and truncation during preprocessing. ### 4. Model Setup * Load `facebook/mbart-large-50-many-to-many-mmt`. * Configure t …

Fine-Tuning mBART-50 for Akkadian-to-English Translation: A Transfer Learning Approach for Low-Resource

Overview The research paper "Fine-Tuning mBART-50 for Akkadian-to-English Translation" by Frank Mor

younes921722/Fine-tuning-gemma3-1b-on-Darija-to-English-translation-task

# Fine-tuning Gemma 3-1B on Darija-to-English Translation Task ## Overview This project demonstrat

Fine-Tuning SeamlessM4T Moroccan Darija to Classical Arabic Speech Translation

Speech Translation From Darija to Modern Standard Arabic: Fine-Tuning Whisper on the Darija-C Corpus Across Six Model Versions

Parameter-Efficient Fine-Tuning for LLM-Based Arabic-to-English Machine Translation

Large Language Models (LLMs) such as GPT-3, BLOOM, BERT... have revolutionized natural language proc

C-EB/Fine-Tuning-mBART-for-Darija-to-English-Translation

Domaine:

Type de record:

Visit

Tasks

Languages

Similaires

Fine-Tuning mBART-50 for Akkadian-to-English Translation: A Transfer Learning Approach for Low-Resource

younes921722/Fine-tuning-gemma3-1b-on-Darija-to-English-translation-task

Fine-Tuning SeamlessM4T Moroccan Darija to Classical Arabic Speech Translation

Speech Translation From Darija to Modern Standard Arabic: Fine-Tuning Whisper on the Darija-C Corpus Across Six Model Versions

Parameter-Efficient Fine-Tuning for LLM-Based Arabic-to-English Machine Translation

Fine-Tuning Whisper for Speech Translation: A Case Study on Translating Darija to Arabic